NLP-Driven Proxy Retrieval for Illiquid Bond Pricing

Arturo Oncevay, Joy Prakash Sain, Chen Ling, Simerjot Kaur, Charese Smiley, Xiaomo Liu, Manuela Veloso

Keywords: illiquid bond pricing, information retrieval, generative ai

Abstract: In finance, pricing illiquid bonds is a complex challenge due to their infrequent trading and limited market data. This paper presents a novel generative AI and NLP-based framework to retrieve liquid proxy bonds, enhancing scalability. Our end-to-end pipeline comprises three modules: (i) Public Information Discovery, (ii) Profiling, and (iii) Matching. Using web data and Large Language Models (LLMs), we generate descriptive summaries and keywords for illiquid bonds and match them with liquid candidates, reducing manual effort. Rigorous evaluation achieved a 71.4% query success rate, and the scalable solution, ~9x faster than a manual approach, has been well-received by industry experts. We are now deploying this pipeline to production, aiming to improve the process of illiquid bond pricing.

Submission Number: 40